Corenel, by Prompd

Digital workers that do the job.

Corenel runs autonomous agents on your machine, under a policy you set: reading, writing and executing with the same guardrails every time, and a full record of what it did.

Start free Watch a run

Free account. Bring your own model key. No credit card.

session · policy: standard

Reads run under the default policy. The write stops and waits for you.

Watch a recorded run →

How it works

Three steps to a working agent.

Run the sidecar
Install the @corenel/sidecar package (it needs Node 22 or later) and start it against the folders you want reachable. Its file tools cannot reach anything outside them.
Pair it with Corenel
Open Corenel in the browser and connect it to the sidecar with the token it prints. Your files stay where they are: there is no upload step.
Set the policy and run
Choose what the agent may do on its own and what needs your approval. Then give it the job and watch it work.

$ corenel-sidecar --root ~/work/api

Runs where your work already lives

Corenel operates directly against your local files, and your shell if you switch it on. There is no upload step and no import: you point it at a directory on the machine you are already working on, and that directory is the context.

How the boundary works

$ corenel-sidecar --root ~/work/api
  listening on ws://127.0.0.1:4858
  root: ~/work/api
  shell off; pass --allow-shell to enable

Policy, not vibes

Every tool call is checked against a policy you control before it runs. The default lets reads run and stops every file write and command for your approval. The same gate applies to Claude Code when Corenel is driving it, so a proxied tool call is governed on the same terms as its own.

Read the security model

Read a file	Runs
Write a file	Asks you first
Run a command	Asks you first, once you turn the shell on
An outside tool that does not say what it does	Treated as a change, so it asks

A full replay of every run

Every agent and workflow run is recorded as it happens: each tool call, each decision, in order. When an agent does something surprising, the answer to “what did it actually do?” is a recording you scrub through, not a log you reconstruct.

See the surfaces

14:02:11 read src/client/retry.ts
14:02:12 grep maxRetries src/
14:02:15 gate write src/client/retry.ts
14:02:23 approved by you
14:02:23 write src/client/retry.ts

One runtime

Six ways to put it to work.

The same policy-gated worker, pointed at a different kind of job.

Agent Crew Craft Workflows Sessions Recordings

Getting started

What an account gets you.

You can watch a recorded run without one. An account is what lets you run your own.

It is free	We do not charge for the runtime and we do not resell tokens. What costs money is the model, billed to you by the provider whose key you brought.
Your work is yours	The workspace, crew and workflows open once you sign in. Run history in your browser is kept per account, so someone else signing in on the same browser does not see yours.
You still bring a model	Add an OpenAI key, point Corenel at any OpenAI-compatible endpoint (including one on your own machine), or drive Claude Code through the sidecar. We do not supply the model, so this step is not optional.

Give it a job.

Create an account, add a key or pair a sidecar, and run. Nothing to migrate and nothing to upload.

Create your free account How it stays contained

Already have one? Open Corenel